error in /var/log/messages: floating-point assist fault at ....

[Sangamesh_B_]

Hi,

A home grown Fortran application (sequential) compiles and executes with producing output on SGI SMP IA64 SLES10 based HPC cluster.
The "top" command shows its using 0% memory and 100% cpu

233511 root 25 0 127m 6528 1952 R 101 0.0 418:19.20 program_exe_name
....

But the /var/log/messages shows following message:

Mar 31 08:59:20 Node kernel: program_exe_name(233511): floating-point assist fault at ip 200000000014ff72, isr 0000020000001001
Mar 31 08:59:20 Node kernel: program_exe_name(233511): floating-point assist fault at ip 200000000014ff72, isr 0000020000001001
Mar 31 08:59:25 Node kernel: program_exe_name(233511): floating-point assist fault at ip 200000000014ff72, isr 0000020000001001
Mar 31 08:59:25 Node kernel: program_exe_name(233511): floating-point assist fault at ip 200000000014ff72, isr 0000020000001001
Mar 31 08:59:25 Node kernel: program_exe_name(233511): floating-point assist fault at ip 200000000014ff72, isr 0000020000001001
Mar 31 08:59:25 Node kernel: program_exe_name(233511): floating-point assist fault at ip 200000000014ff72, isr 0000020000001001
Mar 31 08:59:25 Node kernel: program_exe_name(233511): floating-point assist fault at ip 200000000014ff72, isr 0000020000001001

Please help me to know about this message and does it indicate any harmful effect on the execution process.

Thanks

[Ron_Green]

On Itanium this means that your code is generating IEEE denormal values. These require software emulation to perform.

I found a great write up on this from HP http://h21007.www2.hp.com/portal/site/dspp/menuitem.863c3e4cbcdc3f3515b49c108973a801?ciid=62080055abe021100055abe02110275d6e10RCRD

You can either disable the message OR force the code to flush denormals to zero. Or perhaps you may wish to investigate why your code is generating denormals and perhaps use scaled values to avoid this condition altogether.

ron